MSGNN

This is the repository for paper "MSGNN: Multi-scale Spatio-temporal Graph Neural Network for Epidemic Forecasting", accepted by Data Mining and Knowledge Discovery (DMKD) in 2024.

1. Setup

• Install CUDA 10.1
• run setup_py.sh to install the necessary dependecies for python environments.

2. Usage

To run MSGNN, please follow the steps below:

cd ~/MSGNN/src
# Change directory to the source code folder
python data_utils.py 
# Pulling epidemic data from Google, CSSE. 
python run_models.py --forecast_date 2021-05-01 
# Running ensemble for MSGNN
python run_ensemble.py --forecast_date 2021-05-01
# Get the predicting results
# check '../outputs/2021-05-01_forecast.csv' for details

3. Note

• To run MSGNN, an NVIDIA GPU with at least 6GB memory is required.
• The code is implemented on a server with Intel Core i7 10700F, 32GB of RAM, an NVIDIA RTX 2070 SUPER and Ubuntu 22.04.1 LTS.

4. Citation

If you find this repository useful in your research, please consider citing:

@Article{Qiu2024,
author={Qiu, Mingjie
and Tan, Zhiyi
and Bao, Bing-Kun},
title={MSGNN: Multi-scale Spatio-temporal Graph Neural Network for epidemic forecasting},
journal={Data Mining and Knowledge Discovery},
year={2024},
month={Jul},
day={01},
volume={38},
number={4},
pages={2348-2376},
issn={1573-756X},
doi={10.1007/s10618-024-01035-w},
url={https://doi.org/10.1007/s10618-024-01035-w}
}